Currently recalculating the posterior mode genotype from the called haplotypes is done via brute force i.e. iteration over every possible genotype given the haplotypes and ploidy.
This is efficient to a point but as the number of haplotypes and ploidy increase it would be more efficient to achieve this re-sampling with a simple MCMC.
This resampling MCMC would be mauch simpler (and presumable quicker) than the initial assembly MCMC as it swaps whole haplotypes based on the observed haplotypes so it has a much smaller domain and does not require structural steps.
Ideally we would identify some some threshold where the brute-force approach becomes inefficient and the MCMC approach takes over.
timothymillar
commented
3 years ago
Currently recalculating the posterior mode genotype from the called haplotypes is done via brute force i.e. iteration over every possible genotype given the haplotypes and ploidy. This is efficient to a point but as the number of haplotypes and ploidy increase it would be more efficient to achieve this re-sampling with a simple MCMC. This resampling MCMC would be mauch simpler (and presumable quicker) than the initial assembly MCMC as it swaps whole haplotypes based on the observed haplotypes so it has a much smaller domain and does not require structural steps. Ideally we would identify some some threshold where the brute-force approach becomes inefficient and the MCMC approach takes over.